I think they talked about this being general purpose chip but I would think that Anthropic/OpenAI are at the scale now they could bake LLM weights into chips themselves.
For example, GPT Sol baked into a custom chip run for $100M that runs 10x as fast and 10x as cheap should pay for itself as long as the chip is useful for long enough.
While 2 years ago nothing was useful more than 1 year long, there are many older models in use now (e.g. Haiku 4.5, GPT-OSS 120b), and I expect this trend to continue.
I know this is what Taalas was doing (acquired by AMD), here was their demo, https://chatjimmy.ai/ which is based on Llama 3.1 8B. It feels like this should start to happen soon.
Probably! But not viable yet; the chips would be about a year behind SOTA. Note the ~16 months that the article quotes as being insanely fast to get this chip to tape-out (read: start producing). We'll have to bootstrap our way there: AI is actively being used to get us closer to viable lead times for this.
Unfortunately, there's some real physical constraints: IIRC, manufacturing a wafer takes on the order of a month, start to finish, for the physical processing.
The metal masked ROM is basically only 2 metal/contact layers. It's not a full new design and tapeout. You could roll a new set of parameters every ~2-3months. It's not an architectural change. See statements below.
Part of the key is that by moving even from 6nm to 3-4nm one could embed a 20-30B model as part of a MoE (or only a subset of activated layers) on a single reticle die (note B300s are already multi-reticle), with a separate predictive/dispatch model controlling them each on a separate chip. This is without even stacking CiM ROM die. Moving the layer activations (and KV cache etc) between die requires relatively high speeds (and low latency), but distributed with multiple die in parallel might well be doable even with standard multilane PCIe. Of course KV cache prefill could also be handled by external GPUs. I'm sure AMD will make some reasonable choices.
But that means your different chips all have different sets of weights and are different generations.
If none of that is baked into the chip as now then all the chips are running the latest weights every time.
Even if you could ignore the stuff built into the chip when the time came, at that point you just wasted money on silicon that’s useless in 2-3 months.
Maybe! But it also doesn't require the rate of improvement to slow down. As long as some current model is eventually "good enough" for general use, it could still be a market-killer at a very low marginal price thanks to ASIC. Even if slower, much more expensive models are 10x better, that doesn't actually diminish the utility of the ASIC model, as long as it's "good enough".
How much of that 16mo is design versus just production? If there was a “plug and play” chip where you just BYO weights, how long would it take?
The bigger issue seems to be that these chips can’t hold that many weights at the moment.
(I’m curious if chips with large weights in them would be more tolerant or less to yield issues. If you flip a few bits in the weights, does it really matter at scale?)
Talaas, from what I understand is building stuff just like that. The infra is the same and the weights layer is all you need to change. I guess you could half etch the chips and then finish them with the weights only. I think their turnaround is 6-8 Weeks. The size of the models fitting on the chips at the moment is llama 3 I think?
I know what you're saying, but modulo things like losing track of what year it is as time passes by, a current frontier model is going to continue to be useful for many tasks for many years, even moreso if it's 5-10x faster due to the chip architecture.
It's not that it would be the best forever, it's that it would be useful for plenty long enough to be worthwhile, even if there was better stuff available. In exactly the same way that this computer I'm typing this message on is not the latest and hottest cutting edge stuff. A 7 year old CPU, 7 year old Intel integrated graphics, an older NVMe disk, a mere 32GB of RAM... ok, that's one spec that's still pretty modern although it is slower RAM... but it's still plenty fast enough to comment on HN, even these seven years after it was cutting edge.
exactly, but the "goes out of date" is bad when we talk about software.. but this isnt software, its hardware.
the youd have to buy a new one to get a better model is a FEATURE not a bug.
like if im apple... and i can put a sol level llm in an iphone, market it as privacy first you own your data personal assistant, integrate it all over the os... and then when there is a better model/siri make all the users buy a new phone... thats how they "win" ai.
the old standbys of better screens thinner cameras and batteries arent enough anymore. its basically tapped out. all modern phones are as thin as they need as big as they need as fast as they need and last all day on a battery...
apple needs a new number to up thing that people
can actually feel/see. model generations could be it... every year faster, smarter, more capbilities and integrations.
> but it's still plenty fast enough to comment on HN, even these seven years after it was cutting edge
While it’s still too early to tell, I don’t think that’s how intelligence scales. Better models get you better solutions even to trivial problems. The ceiling for getting it done better is very high even if you’re not doing anything complicated. And difficulty isn’t uniformly distributed anyway - it seems to me that “mostly simple” tasks often have annoying 1% tails that low-intelligence models struggle with. I think we’ll see people chasing the top models for quite a while, or indefinitely - depending on the cost curve.
Google rolled out TPUs in 2015. AWS released Inferentia and Trainium chips in 2020.
If companies working on ML-specific chips was evidence that large transformer models have fully saturated their potential, the field would have been done circa GPT-2.
Yeah this feels like Altman not knowing engineering well enough to realize where the focus should be.
Like he is optimizing to keep providing a vanilla token factory when weighted chips are coming and local models will supplement.
My head canon is savvy chip execs will be etching architecture his OpenAI pioneered into their flagship products while trying to minimize how much foothold he can get in hardware. Murica done offshored it. Not ours to control.
> I know this is what Taalas was doing (acquired by AMD), here was their demo, https://chatjimmy.ai/ which is based on Llama 3.1 8B. It feels like this should start to happen soon.
Taalas needed a giant chip (6nm) for an 8B model.
At best you could use a more advanced node to try to put a MoE model across several chips working together, but you can’t have GPT Sol size models on a single chip like that.
> Taalas needed a giant chip (6nm) for an 8B model.
You're phrasing it like it was kind of an inherent technical limitation with this kind of burning weights into silicon. Which is also not new, it goes back to the 1980s with fixed function digital signal processors and little linear regressions or hardware classifiers for industrial control systems, all are the same basic principle.
It's just usually not worth it to go super small process node, because most models people thought to turn into silicon were pretty small parameter sizes. We're talking 10-100 weight regression or at most 2-4k weight neural net, used in some instrument or factory equipment. You can do a decent MNIST OCR with a 4k weight neural net. For this, 180/130nm is fine.
Or you might think it's required with their special 4-bit as transistor thing (plausible). It's more that when you're experimenting and iterating, TSMC 6nm is their advertised path for rapid prototyping at cost for proof of concepts. And that's already in hot demand, while good luck if you're a startup trying to break in with 3/4nm as your first run.
Nope. But we are hitting some pretty impressive levels with 128B models.
The other thing is, a lot of the time, model performance is improved with more 'thinking' time.
The thinking time is just more tokens... but instead of say 1000 tokens or 10,000 tokens worth of thinking its 1,000,000... how does that improve model performance? Could a 128B model hit levels of GPT Sol?
Yeah, but then there's the size of KV cache needing to be read through that HBM interface for each token, putting a hard limit on the tok/s based on the memory bandwidth.
On some models a large context can be a notable proportion of the size of the weights themselves.
For example, qwen 3.8 27b uses ~64kb/token for the kv cache - so for a 256k token context that's ~16gb of the kv cache for a ~54gb model (assuming 2 bytes-per-param/f16 for both).
So if the current non-baked-in chip is already memory bandwidth bound, as is often the case for current hardware and models, and the "only KV cache in HBM" chip has the same total memory bandwidth, it can only ever be (54/16)=~3.4x faster for the baked in-silicon model.
EDIT: I guess actually (54+16)/16=~4.3x faster, as the current implementation would need to read that KV cache too :)
Yes, they could also sell me GPT Sol 5.6 or 5.7 on a chip and I’d probably buy it. It’s a really really useful model for me, I’m not sure how much better for coding I need it to be. For most things I find Sol good enough with a small amount of coaxing around my tastes.
Keep in mind that what previous work has done on a single chip with weights baked in was on a 8b parameter model. Sol is likely something in the 5T parameter range, perhaps higher. Serving the whole thing at BF16 is on the order of $3m in hardware just to serve it at all, and closer to $1-1.5m of hardware if it was being served as NVFP4. And power draw starting at high tens to low hundreds of kilowatts.
Let's say a magic set of chips comes along to host this. Maybe it's 2-3x more efficient in size and power. You're still talking a form factor that's a good chunk of a rack, draws tens of kilowatts, and could actually be sold at a similar if not higher price point because the OPEX is so much lower.
It may be useful but it's certainly uneconomic to spend >$1m to self host the model, plus ongoing power and maintenance costs, plus the cost to adapt whatever building you're in to be able to power it.
ill give you that the way we talk about this ppl seem to think wed do this tomorrow, but in the 70s a kb of ram took an entire rack and tons of power also. Its seems equally plausible that we could go into a cycle of iterative refinement of baked model hardware that would end up in "personal ai" just like we got to personal computing.
Also right now Sol 5.6 Max is super slow but if it were way faster on a chip (like Taalas' Llama 8b demo) then it would be an extreme value multiplier. But the model is so large that "baking it onto a chip" doesn't seem straightforward.
> For example, GPT Sol baked into a custom chip run for $100M that runs 10x as fast and 10x as cheap should pay for itself as long as the chip is useful for long enough.
but you trade updatability, which I don't think is worth it yet.
Maybe! (1) Would SOL level intelligence be useful 3 years from now? 5 years? (2) would dedicated chips be the most affordable way to run this model in 3-5 years?
I suspect the answer to both of these questions is yes right now, but I agree it’s borderline.
My guess is we only see this once they start saturating computer use benchmarks. That's a use case which would be extremely valuable at the right costs/speed, but the current models just aren't there yet.
"Baking in" a model into a chip is a bad idea because chips take 2 years to tape out and then you're stuck doing inference on llama 3 in 2026 when fable/sol are available. Every accelerator is a tradeoff between flexibility and performance and GPUs are already pareto-optimal
That's only half the problem. OpenAI is contractually obligated, if you will, to believe that models will continue improving at an impressive rate for the foreseeable future (otherwise their valuation makes no sense).
If you believe that, then you should expect to get Sol-level performance out of a Luna-cost model within six months or a year. If you have a system with the weights baked in, that means you're going to end up serving that Sol-class model several times more expensively than it will take someone who comes along a few months later. (such as what recently happened with DeepSeek's update.)
And under that assumption of continuing advancement, baking things in doesn't make sense in general - it's a play you'd make if you think things are slowing down a lot. Which may be right but it's not OpenAI or anthropic's play.
Assume a $800 billion valuation. $100 billion ad network. $30 billion op income. 26x price to op income ratio. It's right there for them to grab, or someone else to grab.
Their valuation does make sense if you believe: 1) they can retain a massive user base and 2) a massive user base can be monetized. Future value is almost always pulled forward these days for high growth tech companies.
An LLM the size of Google search in users is even more valuable than Google search. The ad market for LLMs will be even larger than search was (no matter what HN prefers).
The monetization part is the easier part. Silicon Valley understands extraordinarily well how to build ad networks. If OpenAI maintain their gigantic user base, a $100 billion ad network is a given bolt-on. They'd have to screw that up in an epic way to not get there.
Facebook - Insta - WhatsApp is an absolute dogshit tandem with a gigantic user base. $228 billion in ad sales and still expanding 10% per year.
Google knows this is what's happening, that's why they don't care about chasing Anthropic very much. They're busy completely remaking how their core search business works.
Good enough will hit when the tech stops advancing quickly. You could have a "good enough" model but in 2 years if the general purpose chip can run it just as fast, there is no point having the single purpose one.
The whole point is that it's supposed to be more efficient. But models are also still getting absurdly more efficient every year, so you're likely nullifying much/most of the advantage. 18 months is a long time right now (and 18 is only time to tape out, not operational in data centers).
Even if the balance was net positive, you would also not be able to train them against new tools/harnesses or knowledge. How many years do you expect to keep using them?
The good enough level isn't ever arriving. We're in the first or second inning for LLMs. They will rapidly subdivide in complexity, they will not stagnate in the next decade.
Beyond the model, when would you freeze processor performance, such that it was good enough? Because that's exactly what freezing on Talaas is premised around.
The semiconductor technology will also continue to improve. You lose twice. Talaas is one of the dumbest ideas I've seen in semiconductors in decades.
There are other options. I worked for a startup called NVXL and we were programming DNNs into FPGAs using OpenCL, on custom boards we built to plug into NVME. It worked great, but it wasn't fast enough at the time to compete with Nvidia, or even Intel AVX512. Ultimately the company failed, but maybe some hybrid like that could work for LLMs? I haven't been up to date on how DNNs and LLMs look like under the hood these days, but there's got to be someone doing something similar.
I've seen several attempts even on HN of the LLM meets Teddy Ruxpin (or more accurately AG Talking Bear) but most of them offloaded the AI to some off-site servers.
I’d like to think that most parents would be weary of handing their children what basically amounts to a tape recorder that siphons all the data off to a large corporation.
OTOH, a completely local one (LLM + VAD + Speech Rec) would be a fun little thing to build.
These nascent inference chip efforts are reminding me of the early 3dfx / riva / mach / powervr days. Will be interesting to see if inference chips are here to stay and, if so, who the eventual dominant player(s) will be
Which in turn reminds me of Soundblaster audio cards! I suspect inference chips are closer to the GPU story than the Soundblaster story though.
I remember one soundblaster card I bought came with a Lara Croft demo, that exploited the incredible immersion of real time dynamic reverb.
Genuinely I think game audio took a few steps back from that heady era, the innovation in audio likely didn't sell as many cards as graphics innovations did.
EAX was very powerful in its heyday, but it has died because of a thousand cuts.
First we had to have the audio processor. Good EAX was available on top of the line cards, and they were not always cheap. Lower end chips got less features.
Then we had to have the speaker setup to have the greatest sound, or needed to get a real 5.1 headphones, which were bulky and never provided the same fidelity.
Then Microsoft changed the Windows driver model, cutting the driver's direct access to the card. All of the timing sensitive effects were gone in an instant. I remember installing the new drivers and getting literally nothing. Sound Blaster was the only card with an hardware mixer, and Microsoft didn't feel like enabling them. Mixing at the DirectX layer killed the cards.
Soundblaster's very closed stance didn't help them either. None of the cards after Audigy2 worked with Linux when I had my desktop system.
After my Audigy2ZS, I moved to Asus Xonar D2X. Its positional audio capabilities were nice, but I mostly bought it for its Linux support and sound quality, and that was top notch in that regards.
Then sound cards became commodity. Everybody stopped making good cards. Musicians moved to audio interfaces, audiophiles moved to DACs.
Just looked to the SoundBlaster website. Internal cards are very limited. One DAC, one DTS enabled 7.1 sound card for PC cinema systems, three game oriented lower end cards, nothing else.
Eeeeeh idk about the Sound Blaster comparison. Creative earned their place in the early-mid 90s solely because they were the one company making a sound card with drivers that actually worked properly.
It wasn't really the cool reverb effects or wave tables, though those were a nice bonus. It was just "I can tell my computer to make sound and it actually makes sound without days of troubleshooting."
Granted, similar things could be said about 3dfx. It's was a 3D card with drivers that actually worked.
And then there's the obvious "sound blasters and voodoos go in my computer, jalapeno goes in someone else's computer" thing.
There is drastically more power and profit in the software ultimately.
Apple is in the software first, the hardware second. Everyone at Apple has been trained to understand this for decades, and Jobs pointed it out endlessly. Apple's real moat is software (services, iOS, experience, MacOS).
Windows, Office, Azure, et al. Microsoft accumulated approximately one zillion dollars in profit on the back of software. It's a vastly superior business to anything hardware has traditionally seen. Nvidia is the first true juggernaut hardware profit machine, and the AI boom in extended hardware (RAM, storage) will prove temporary (even if there is a feast during that time). Microsoft's advantage and moat was Windows-Office for decades. It was a far better business than Intel's chip biz.
Google is a software company first. Every aspect of what made them and maintains them is software first, hardware second. They're a $400 billion software company. Their ad machine is software. Search is software.
Facebook is software. Instagram is software. WhatsApp is software. A $200 billion software company. They're not selling hardware, they're selling ads via software, they're monetizing users that use their software.
AWS is at least half software as an entity in terms of complexity, competitive advantage, et al. That's a two trillion dollar business.
LLMs can run successfully with various hardware approaches. The software is the value at the end of this, regardless of the hardware under it. The sole exception so far that may be sustainable is Nvidia, and we'll see if the bottom falls out from under that margin monster (China, specialized AI chips, whatever it happens to be that cuts under them massively).
Hardware always gets its margin squeezed eventually because it's a manufactured good (with inventory, fabs, etc). Software is hyper margin by default, you have to layer a lot of garbage on top of it to kill the margin. Nvidia is 33 years old, they have had a rich business for three years, that's it.
The AI boom is the sole reason anything in hardware has looked great in the past 20 years. Check the margins & op income for the top 20 hardware companies, from TI to AMD to Intel to Nvidia to Micron to Sandisk to Samsung to TSMC to ASML, prior to the AI boom of the past couple years. It won't last indefinitely. And after the return to a more normal environment happens, the hyper margins in software will persist.
> Will be interesting to see if inference chips are here to stay
To me, the efficiency gains of inference chips are so significant that they are certainly here to stay — barring a revolution of sorts that leads to a world devoid of AI as we know it.
This couldn't have been easy. The team at OpenAI has worked a miracle.
For example, Meta and Microsoft’s AI ASIC programs not getting off the ground despite being at it for much longer shows that cost is only one part of the equation.
I bet cost is of no issue with the capx where it is at. It is almost certainly organizational. Meta throws money at every problem and it never seems to workout for them.
It's so funny to see FP4.... I remember 20 years ago being asked what sort of HPC we needed in genomics, and the answer was basically, "lower precision, faster" for the stuff I was working on. But FP4 is, well, almost comical.
One thing not on that comparison table: die size. If I'm understanding that correctly, it's about the same as the Rubin, but at 1/3 the number of NVFP4 PFLOPs. (The text disagrees with the table, I'm taking the table as truth, perhaps that's wrong...)
Agree, I remember when even half precision made its way into C# sometime around 2020 (I didn’t know much about ML then) and I thought, well I guess that’s a worthwhile tradeoff but I can’t imagine going lower. Lo and behold (1-bit Bonsai) how much lower you could go.
They were right about everything 50+ years ago, but they didn't have the budget for the right hardware, had to write conference papers and books instead.
I can totally see how ternary would work from a physical implementation perspective but I have a really hard time visualizing anything using base-e, can you explain how such a thing would work in practice?
I hadn't seen the token/Joules comparison with human speech before. Humans are still 22x more efficient, which is not that far considering the rate of progress in this area.
The 20W number includes EVERYTHING else the brain does. The chips/models are literally only producing tokens. Let's see an LLM drive a robot harness and have the robot produce speech, as well as move through 3D space, keep track of metabolic needs, etc. etc. etc. before we compare efficiencies. That is even assuming the tokens are of equal quality. This comparison is currently Apples and Oranges.
And the brain is literally only producing electrochemical signals.
I don’t see how tokens can’t produce speech or track metabolic needs. You can talk to chatgpt can’t you? Or do you mean literally talking? Because that’s not a brain function, that’s the mouth, vocal chords, and lungs.
> I don’t see how tokens can’t produce speech or track metabolic needs.
It probably could, but the point is this would require additional tokens, blowing up the comparison. The token output of LLMs and "token output" of speech are simply at different abstraction levels. Hence my comparison to the LLM brain driving the robot harness to produce speech etc. This would be more comparable, and also look significantly worse than "only" the 22x less efficient number.
22 times more efficient is not like 22 times more powerful. It’s extremely harder to close the gap in power efficiency than in raw power. Simply because the power efficiency we see is the result of billion years evolution optimization.
But the true number is IMO far bigger: orders of magnitude greater if we think in terms of equivalent performance.
I couldn't source the parameters from the screenshot or the nearby graphs, but from the nearby graphs you can see that at concurrency C=1, tokens/Joule (vertical axis) has totally plummeted, and obviously concurrent inference is much more efficient by batching. Divide the memory by the bandwidth and thats how long it takes to dump the full RAM contents through the chip. Do you want to do this once per token for a single conversation, or do you want to progress multiple conversations if you're going through all the weights anyway? The peak in the graphs is easily 22x more efficient than the low bottom right part on the graphs. So in batched mode its already more efficient than human speech.
I do believe that this is the trade off. We are more efficient but slower in terms of thinking (at the same level of intelligence). Some animals go much further in terms of that trade off, see https://en.wikipedia.org/wiki/Portia_(spider) for example.
Just checked wikipedia page... they have like 100K neurons only.. wtf!!! How can nature cramp all senses, including spatial, motion, life maintenance and general thinking into just 100K neurons?
The claim that tok/s independent of quality is a useful comparison (I can get thousands of tok/s on a suitable small model), and secondarily that humans can’t output “tokens” faster than than in some sense, which I am less confident about
At just inference! Which both a human and a model can not do without training, but while training rounds to zero for the model, for humans it scales linearly.
I am relatively certain we have already squarely been beaten in net efficiency at scale.
I love how now you have to consider the possible s** posting motivation behind analysis of a trillion dollar industry being conducted at a world-class level by a bunch of ex Reddit and 4Chan adjacent mods -- it's one of the best stories in AI that SemiAnalysis is not cut from the same cloth as Gartner McKinsey et al
The guy that was part of FTX, fired from openai for alleged theft, got billions in a hedge fund somehow then lost billions. Why are all these people so scummy? It is like voting Trump three times in a row.
They're very intelligent people who do very clever things at a young age, which draws the attention of very rich people who can exploit them to get richer, and no one tells the young person they're being exploited. They're heaped with praise and 'wealth' (millions, but crumbs compared to what they're making for other people), and told they're geniuses who can do no wrong, mostly by the media that happens to be owned by the rich.
Then the rich people pull the rug leaving them holding the bag, and they move on to the next young clever group.
The "industry news and research" part of the AI industry feels very... suspect to me. My intuition is telling me that it's a bunch of people with influencer-y type social media skills and no actual credentials just grifting because there's so much money floating around.
The semianalysis people have scripts which incorrectly count their numerators and denominators all the time. All their benchmarks are flawed. It is such a slipshod operation and they charge exorbitant amounts of money for it.
> I love how now you have to consider the possible s*** posting motivation behind analysis of a trillion dollar industry being conducted at a world-class level by a bunch of ex Reddit and 4Chan adjacent mods
I mean, previously you could have said something much the same except substitute "frat boys".
In short, better hardware will drive down token cost in the near-term, but will drive up the demand for tokens as it gets cheap enough for other sectors to start to use it heavily.
It comes from steam engines where economists originally thought that coal demand would plummet with more efficient engines, but it actually just meant that we found more uses for steam engines.
If we are applying Jevons paradox to this then the unit being consumed is not tokens but the inputs for token production - power, capex, something else. To draw an analogy to the steam engine, coal:electricity::mechanical-work:tokens. Jevons paradox does not talk about mechanical work becoming cheaper in the short term setting up a sort of rubber band of demand creating spiking prices for mechanical work. Compared to the renaissance, mechanical work was much cheaper throughout the industrial revolution and remains cheaper to this day. We can still definitely say that the easier it is to produce tokens, the cheaper they will be.
All Jevon’s paradox says is that as a resource becomes cheaper total consumption of that resource increases. It applies equally well to the inputs of token production as it does to the tokens themselves. The former would describe the effect the sellers into AI companies see (energy, GPU chips, RAM etc - if they lower their prices they’ll have more overall consumption) while the latter describes what the AI companies see with their customers (if they lower token prices consumers will use more tokens overall).
I would posit there’s no way in hell they’re getting sufficiently cheaper on a short enough time frame vs how much demand is sky rocketing. AI companies are seeing quarterly doubling of revenue if not more.
That's when demand is higher than capacity. Now imagine places like Gigalab and Chinese labs are online and able to produce significant percentage of chips. That could cause real surge in prices.
I think you're reducing a very complex thing (the global economy) into a very simplistic model (Jevons' paradox) and thinking both are the same thing. This has no predictive power or rigor. You're just wishing things would happen as they did before, without considering that conditions and situations change significantly, and instead of Jevon's paradox, we look back at today 50 years from now and talk about Jensen's paradox.
This doesn't mean the concept is BS, but one single concept cannot explain away everything in such a system.
There is just so much downward pressure on token price, from every direction. We would need a completely new understanding of economics to explain why the price shouldn’t go down. Or market collusion/regulatory manipulation.
The demand for them is growing _per person_, not just across the wider economy, if tokens cost half as much but you want to use 3 times as much you're going to have to pay more.
Maybe 1000s of tokens per second unlocks realtime robotic decision making, and now every robot needs to continuously stream tokens to and from the cloud to operate. That could 1000x demand overnight, just to speculate :)
I would very much like it if anything that moves with appreciable mass is governed locally just in case the link drops and/or latency suddenly goes up. Motion is very unforgiving and accidents will happen if that's not taken into account.
I mean, it will obviously crash at some point. With so much pressure on token price to go down that means way less opportunity for margin for AI providers. OpenAI is in a pretty bad situation
With the corollary that old hardware valuations will plummet with them.
Although given we have marginal pricing we need to push through to those lower prices in the face of increasing demand, so timing of this is uncertain and the key to the AI financial markets
Hopefully this also means billionaires can stop trying to drop data centers into residential neighborhoods with zero noise control and polluting on-site generators, signing local politicians on with NDAs, calling for eminent domain to seize homes to build power lines to data centers, etc. etc. etc. Not to mention the water use controversy.
Token prices plummeting is probably a good thing, but not without the regulatory backstops that prevent these effectively industrial facilities from being operated with no regard for the externalities they impose on people who live near them.
Nah, Jevon’s Paradox says that cheaper tokens will mean increased overall energy consumption.
If we can’t even build data centers, the least disruptive industrial use possible, there’s no hope to reindustrialize the US or anywhere outside of China.
this is a story about a proprietary accelerator being built/designed by a token provider. and you think they're going to return the efficiency gains to the customer instead of capture the value for themselves? interesting take.
GPUs and memory have gone up in price. It's more expensive to buy a 1-2 year old video card than it was at launch, sometimes by a shockingly large factor. Laptop vendors have recently shipped flagship models with less memory than the previous model, because they can't match price expectations for a laptop.
i figured out why this comment is so confusing: this is actually a message from the past, around 2020. either that or simianwords is a time traveler that arrived today and hasn't read the news yet.
We should be mindful of the context that many of these providers VERY likely have been selling their subscriptions at a substantial loss
So as much as i agree “more profits to stakeholders screw the customer”, i think its more of an emergency to get to profitability before the music stops.
He's a subscription truther. There's loads of them. OpenAI's profit increases with each subscription that is cancelled. Pretty soon they'll have more profit than God.
This is most impressive. The interesting question to me, is outside of the LLM accelerator space: will generalized chips have massive leaps in performance once LLM technology is used to create the next generation? In general, will we see rapid advances while we extract the value of these models in creating architectures? I'm so far removed from the space that this is a very naive interpretation of all this, but I'm curious.
ASICs always do better than general purpose chips. General purpose chips is turtles and turtles of virtualization and have to consider 4+ decades of backward compatible instructions set support.
ASICs are deployed when the application area is economically large enough to so there is return on the investment in developing one. Bitcoin mining few years ago or today inference or more mundane things like video decoding/encoding.
General purpose chips on the other hand have to be general purpose first to be useful, i.e. support as many application domains and instruction sets as possible . It can be long tail of support which both slow your chip down and also slow development down. Apple's took a long time to develop M series to be general purpose enough and still need even now software tooling like Rosetta to make say virtualization work for a good reason.
New tooling would always help and there is already lot of software emulation for developing chips today but you still need physical iterations to tap-out and have high enough yield, no LLM can help with that.
Existing CPUs have been extremely optimized by ~6 competing, well-funded teams. I expect AI to accelerate things somewhat but it's not clear that there is any low-hanging fruit available for AI to find.
The reliance on Deepseek and Kimi as the benchmarks from every chip maker from NVIDIA to OpenAI is a good tell of where things are heading. In the next couple of years, hopefully we will have systems at home for everyday use and corporations can buy bulk from providers.
As opposed to closed-source models? Benchmarks for GPT Sol wouldn’t be particularly meaningful, as no one else can run the benchmark, and we don’t know what the exact model specs are.
Picking the best open source models is really the best they can do.
If the populist campaign is to Make Affordable DRAM Again, then it's not a terrible solution.
The current datacenter owners love a compute-bound world anyhow. A moratorium on new datacenters would increase their valuation, encourage efficiency and make computers cheap again. If Chinese labs can ship frontier models under 1T parameters, why not American labs too?
If you think tanking Trillions in investments, warming the earth and increasion ocean water levels, creating water shortages and brown-outs is "good for all of us" - well, the rest of us beg to differ.
these GPUs make computation faster, I understand as of now maybe all the computation is used to generate yet another junk LinkedIn post or unnecessary RFC, but at some point this craze should settle and we will be left with powerful computation machines, which can be used for computing more useful things
The GPUs being paid for w/ billions in investment will be obsolete and e-waste in a few short years same as a Cray-2 was just a decade after its release.
It's fine if you're one of the people selling shovels to gold miners for a while, but sucks to be building houses in the boom town?
I hope the LLM wave will leave GPUs behind to go back to pursue more general-purpose computation rather than spending their die area on multiplying 4-bit-number matrices and such things.
When people talk about the commodification of inferencing, they imagine a future where everyone has access to frontier models and can run them at the same cost, and what will actually happen is closer to the commodification of _oil_, where only a few companies have the scale to produce it at a competitive price, and advances like this are _why_.
Once models are more or less interchangeable, the price of LLMs will drop to essentially the price of energy required to run them, and the big labs will be able to run them cheaper than anyone else.
I don't agree. At the moment companies like NVIDIA take several times what it costs to make a chip. I think the fair split for the technology contribution is more like 50-50, maybe even 30-70 in favour of the manufacturer.
With competition we will actually have the fair split, whatever that is, and thus much lower prices.
At the moment, to have a big AI firm, or really AI firm at all, you need to be blessed by NVIDIA, in the form of receiving circular financing for your compute. They know that their prices aren't fair, or competitive.
Commoditization of inference is the end of that. The end of the mega-premium on inference hardware, and it's good not only for people who like running their LLMs, but it's the first step towards commoditization of training.
I don't believe models will be commodified because each model is unique with strengths and weaknesses. Its not like Steel which is more or less the same no matter where you purchase it from.
If what you said were true, you would hardly see people complaining about the quality of Opus 5 or good writing from Sol. But people do.
The same level of intelligence gets roughly 10x cheaper per year. So you might both be correct where a large part are commodity tasks but frontier is hard and valuable and not commodities.
This depends heavily on what the use-case is. Yes, if it's a coder making software and having to read LLM output then writing style matters. If the LLM is used in an automated data processing pipeline with a capped level of complexity, entirely different aspects matter and LLMs become more interchangeable.
> Its not like Steel which is more or less the same no matter where you purchase it from.
I’m not an expert in metallurgy by any means, but this seems really off. There are many recipes for steel and varied processes that also impact the final product.
> However, as previously mentioned, Jalapeño’s results are obtained without speculative decoding and Vera Rubin’s results use speculative decoding. Speculative decoding leads to a ~3-5x reduction in cost per token. When speculative decoding is implemented on Jalapeño, this will enable Jalapeño to serve tokens even more cost effectively.
How much speculative decoding improves throughput is workload-dependent. Yes, it can improve performance by 5x, but it can also slow down performance by 2x. It also requires a vastly more complex decode loop than the standard token-by-token decode. The tech journos didn't ask themselves if speculative decoding improves perf so much why wasn't it on by default? Perhaps, because speculative decoding is not optimal for OpenAI's ASIC? Perhaps that is also why they were only able to benchmark the not-very-representative single-user-decode case?
I was initially just surprised by how bullish Dylan is on OpenAI/Anthropic and how bearish he is on China, despite Chinese labs getting closer to US SOTA while offering inference at dramatically lower prices.
So, I started digging while waiting for various day-job inference calls to return, ha.
Dylan says he spent years obsessively posting on hardware forums, moderating hardware subreddits, and running anonymous hardware blogs/videos before SemiAnalysis. But he also says most of that history is now gone, including from the Internet Archive, because he asked for it to be removed.[1]
In a 2024 interview he described his post-college job as “data science” around hurricane/earthquake/wildfire simulations for a financial company.[1] In a 2026 Sequoia interview he described himself as having been a “quant at a small quant risk firm” who generated $10M+ of “risk-free revenue.”[2] The Information reports that he declined to identify the employer and doesn’t list it on LinkedIn.[3]
Even harmless/silly stuff seems to drift. In February he said he kept bees for ~1.5 years. Today it was “few months, few months.”[4][5] I know, sort of silly and doesn't matter.
The Information reports that Patel owns stakes in ~20 startups in the same ecosystem SemiAnalysis covers, organized a $50M Fluidstack SPV, and is now targeting a $400M venture fund.[3][6]
And, in a 2022 HN discussion about SemiAnalysis disclosures, after saying his reports had moved smaller stocks by 20% in a day, Patel wrote: “If I thought I could move the stock, I'd make the position in the morning alongside my clients, and publish shortly after.”[7]
I don’t know that any of this is false or that anything improper happened (I’m definitely not claiming that). More that 1-2 of these things would just be odd. Taken together, though, they made me question how much trust I was putting in the broader story.
The dynamic of reminds me of crypto, WeWork, Theranos, Citron, etc. Once enough important people validate someone, things that would normally invite basic diligence somehow stop getting questioned.
Not publicly acknowledging how misallocation of capital may be happening today shows he is corrupt. He’s not that dumb to not know it’s a major risk to the whole story, and is certainly financially incentivised to write as he does.
If you research the origins of Dwarkesh, even more conspiracy level points emerge. I do not think even in the handful videos post Leopold fund collapse he has addressed it in any ways. That is the point of so called observers, they pretend to be impartial but everyone is a hustler in some way.
Once again we see the classic PR hype machine tactic of comparing a newer chip which is only available as an engineering sample to other chip designs which are widely available and much older.
They also fawn over the chip’s TDP when all other chips have to support 16 bit floating point and thus must run much hotter.
They make the classic mistake of equating max TDP with in-use-watts, and praise this magnificent (fictitious) performance per watt at FP8 with other chips’ max-TDP at FP16, which draw twice the power.
Story says they're power limited. That's half-true. Actually they're water-limited. To generate power, you need water. To cool chips, you need water. If you try to use less water on one side, you need more water on the other side (it's physics ya'll, making and using energy generates heat which requires dissipation). The world's freshwater is diminishing while also being consumed at an alarming rate. The future AI oligarchs are whoever controls the most water.
The other side of the conversation is the idea that large models in DCs on custom silicon is the future. Maybe for enterprise? But consumers will eventually (10 yrs) have affordable hardware designed to run crazy-good local models (more RAM + higher bandwidth). That will take pressure off of datacenters, but also reduce AI profits, and move that money to consumer chip/device makers. Apple is once again the biggest winner. Nvidia consumer chips might get cheaper, but nerfed, to encourage datacenter use where they make more money. I'm hoping AMD can stop being terrible at software so that when we finally have their better hardware we can actually use it.
For datacenters specifically I've never understood what specifically consumes the water. Arent the water-cooling loops closed, so the water just cycles around and around and around?
The system which runs coolant over the chips can be closed but the part which uses an evaporative system to cool that is still open loop and vents water into the air, no?
Evaporated water is condensed, and in the process transfers its heat into another place that removes it. Another simple example is a pot of boiling water with a lid on it.
At the datacenter side, it depends on the method of cooling. You can chill the air or the chips directly (or both), doesn't matter, you still need to cool, and that still needs water. The question is, where is the water being used?
- If they use either evaporative cooling or a liquid-cooled heat exchanger, that uses tons of water consistently. This requires less energy (it's mostly passive) so you use more water.
- If they use closed-loop water cooling and/or heat pumps/electric chillers, that uses much less water - at the DC. But it does require more energy to circulate the water, run fans, etc. If you are using more energy, where is the energy coming from? It's coming from power plants, which require... you guessed it... more water (e.g. thermoelectric, hydroelectric, geothermal, concentrated solar). They need water in order to generate the power, and lots of it. Coal, natural gas, nuclear, and concentrated solar, all use steam to generate energy. Nuclear also uses water to cool the reactor. And water is used extensively to extract coal, oil, and natural gas. Geothermal uses water in the ground.
You can't not use a ton of water in one fashion or another. It just depends what method, and on what end the water is used. And the crazy thing is, most new datacenters are being built in places with extremely little water. Guess how that's gonna work out as the planet gets hotter?
I don't know why I got downvoted to hell for stating facts every datacenter architect knows. HN be HN'in.
How can OpenAI mass produce this chip at scale more economically than Nvidia which has experience in the supply chain and scale efficiencies to do it efficiently?
NVidia has enormous operating margins, so a competitive solution doesn't have to match or beat NVidia's scale efficiencies; it just has to beat delivered cost.
One objective of the project might be simply to provide credible negotiating leverage when dealing with existing suppliers like NVidia. You don't have to deploy at scale for that to work, but you do have to look like you could if pushed hard enough.
> NVidia has enormous operating margins, so a competitive solution doesn't have to match or beat NVidia's scale efficiencies; it just has to beat delivered cost.
But then that means you have no actual moat against the behemot, right? Your competitor can move into the market as soon as they want to, at much better cost (so at slightly better price)... and Nvidia certainly can adapt much faster around hard hardware specs innovation than a new entrant ever could.
In the short and medium term, it probably won't be more economical to produce for OpenAI. Where OpenAI is benefitting from their own chip is being able to tailor it to their models and workloads. When you buy off the shelf Nvidia, its not perfectly tailored and OpenAI has to spend marginally more to run off that chip. At the scale OpenAI is operating at and plans to operate at, that margin becomes pretty big $$
"How can OpenAI produce a LLM at scale more economically than Google, Amazon, and Microsoft which have experience in planet scale software and scale efficiencies unlike them".
One answer is they're quite good at poaching talent.
Nvidia buys the memory it uses on its GPUs, same as all other ASICs.
To give some context, Intel started making DRAM, I think they were actually the company that came up with modern memory techniques. They exited the market and pursued a more lucrative moat with CPUs.
People keep saying stuff like this without understanding what it takes to make RAM. It's one of, if not the most, heavily patented things in the world. The second you dip your toes into those waters the lawsuits begin.
If somehow you get around the patent issues, you're now faced with huge research and development costs, fabs to build, processes to sort out and all of that has very high failure rates.
Last time I checked Micron was the largest patent holder in the world and even for them this is a hard area where they are number 3 in the market.
RAM chips are not hard to produce compared to many other types of semiconductors; Intel started in the memory game and left because the margins weren't great and they were going to fold. The failure rates on these chips are actually very tolerable; you can have a very bad yield and still have a viable chip due to things like ECC.
Well Sam Altman finally has built a moat against Chinese open weight AI. Well done. But what will this mean for Cerebras?
I remember when Tesla was building its own inference chips, and after about 2 years and billions spent, the whole effort was scuttled b/c they simply could not keep up with the iteration and R&D cycles of dedicated chip companies. I suspect the same will be the case with OpenAI vs Cerebras + Nvidia/Groq
Cerebras is targeting a distinctly different point on the cost/latency curve. They are betting that there will be some high value applications where latency and not just throughput is super important.
It is being used as part of a combined system. For example AWS is pushing for Trainium + WSE 3. The WSE 3 does the decode and the Trainium does the prefill.
Even in nvidia land rubin + LPU does a similar thing.
It has its downsides of course - if your traffic swings prefill heavy to decode heavy, you can't suddenly use your lpu for prefill. With GPUs they're totally interchangeable. Tradeoffs.
> Well Sam Altman finally has built a moat against Chinese open weight AI
Hes got a press release.
The issue is, baking something to silicon requires discipline and about 2 years.
This isn't something you can just change your mind on halfway through. Trust me, I know. You need a clear vision of what you want to support, why and what bits of a chip you need to achieve that.
I think the Chinese are going to be building their own chips aided with AI. DeepSeek, z.AI, MiniMax, Moonshot, etc, it's a race. The take off has really started.
> and after about 2 years and billions spent, the whole effort was scuttled b/c they simply could not keep up with the iteration and R&D cycles of dedicated chip companies
That sounds quite like...nonsense?
Chip companies work on years-long cycles. They know today what are they launching 4-5 years from now.
I think they talked about this being general purpose chip but I would think that Anthropic/OpenAI are at the scale now they could bake LLM weights into chips themselves.
For example, GPT Sol baked into a custom chip run for $100M that runs 10x as fast and 10x as cheap should pay for itself as long as the chip is useful for long enough.
While 2 years ago nothing was useful more than 1 year long, there are many older models in use now (e.g. Haiku 4.5, GPT-OSS 120b), and I expect this trend to continue.
I know this is what Taalas was doing (acquired by AMD), here was their demo, https://chatjimmy.ai/ which is based on Llama 3.1 8B. It feels like this should start to happen soon.
In case anyone's interested in these niche startups like taalas, here are a few more:
1. https://matx.com/
2. https://www.d-matrix.ai/
3. https://www.etched.com/
4. https://www.positron.ai/
5. https://hyperaccel.ai/
6. https://axelera.ai/
7. https://www.enchargeai.com/
8. https://furiosa.ai/
also https://velaura.ai/
Probably! But not viable yet; the chips would be about a year behind SOTA. Note the ~16 months that the article quotes as being insanely fast to get this chip to tape-out (read: start producing). We'll have to bootstrap our way there: AI is actively being used to get us closer to viable lead times for this.
Unfortunately, there's some real physical constraints: IIRC, manufacturing a wafer takes on the order of a month, start to finish, for the physical processing.
Maybe once LLM improvements asymptote further?
The metal masked ROM is basically only 2 metal/contact layers. It's not a full new design and tapeout. You could roll a new set of parameters every ~2-3months. It's not an architectural change. See statements below.
https://www.eetimes.com/taalas-specializes-to-extremes-for-e...
https://www.turingpost.com/p/taalas
https://cambrian-ai.com/taalas-launches-hardcore-chip-with-i...
Part of the key is that by moving even from 6nm to 3-4nm one could embed a 20-30B model as part of a MoE (or only a subset of activated layers) on a single reticle die (note B300s are already multi-reticle), with a separate predictive/dispatch model controlling them each on a separate chip. This is without even stacking CiM ROM die. Moving the layer activations (and KV cache etc) between die requires relatively high speeds (and low latency), but distributed with multiple die in parallel might well be doable even with standard multilane PCIe. Of course KV cache prefill could also be handled by external GPUs. I'm sure AMD will make some reasonable choices.
But that means your different chips all have different sets of weights and are different generations.
If none of that is baked into the chip as now then all the chips are running the latest weights every time.
Even if you could ignore the stuff built into the chip when the time came, at that point you just wasted money on silicon that’s useless in 2-3 months.
> Maybe once LLM improvements asymptote further?
Maybe! But it also doesn't require the rate of improvement to slow down. As long as some current model is eventually "good enough" for general use, it could still be a market-killer at a very low marginal price thanks to ASIC. Even if slower, much more expensive models are 10x better, that doesn't actually diminish the utility of the ASIC model, as long as it's "good enough".
They could etch the model architecture, without the weights into the chip.
This way newly post-trained model can be loaded and served the same day.
How much of that 16mo is design versus just production? If there was a “plug and play” chip where you just BYO weights, how long would it take?
The bigger issue seems to be that these chips can’t hold that many weights at the moment.
(I’m curious if chips with large weights in them would be more tolerant or less to yield issues. If you flip a few bits in the weights, does it really matter at scale?)
Talaas, from what I understand is building stuff just like that. The infra is the same and the weights layer is all you need to change. I guess you could half etch the chips and then finish them with the weights only. I think their turnaround is 6-8 Weeks. The size of the models fitting on the chips at the moment is llama 3 I think?
> I guess you could half etch the chips and then finish them with the weights only.
Basically a https://en.wikipedia.org/wiki/Gate_array. (The non-field-programmable kind.)
tapeout could shrink but days per mask layer (DPML) does not have much margin..
I'm not in industry, is DPML (which I assume is the time required to make a mask?) set by electron beam scan time or something?
I think Sol is already good enough though.
"640k (token context) should be enough for anyone."
I know what you're saying, but modulo things like losing track of what year it is as time passes by, a current frontier model is going to continue to be useful for many tasks for many years, even moreso if it's 5-10x faster due to the chip architecture.
It's not that it would be the best forever, it's that it would be useful for plenty long enough to be worthwhile, even if there was better stuff available. In exactly the same way that this computer I'm typing this message on is not the latest and hottest cutting edge stuff. A 7 year old CPU, 7 year old Intel integrated graphics, an older NVMe disk, a mere 32GB of RAM... ok, that's one spec that's still pretty modern although it is slower RAM... but it's still plenty fast enough to comment on HN, even these seven years after it was cutting edge.
exactly, but the "goes out of date" is bad when we talk about software.. but this isnt software, its hardware.
the youd have to buy a new one to get a better model is a FEATURE not a bug.
like if im apple... and i can put a sol level llm in an iphone, market it as privacy first you own your data personal assistant, integrate it all over the os... and then when there is a better model/siri make all the users buy a new phone... thats how they "win" ai.
the old standbys of better screens thinner cameras and batteries arent enough anymore. its basically tapped out. all modern phones are as thin as they need as big as they need as fast as they need and last all day on a battery...
apple needs a new number to up thing that people can actually feel/see. model generations could be it... every year faster, smarter, more capbilities and integrations.
> but it's still plenty fast enough to comment on HN, even these seven years after it was cutting edge
While it’s still too early to tell, I don’t think that’s how intelligence scales. Better models get you better solutions even to trivial problems. The ceiling for getting it done better is very high even if you’re not doing anything complicated. And difficulty isn’t uniformly distributed anyway - it seems to me that “mostly simple” tasks often have annoying 1% tails that low-intelligence models struggle with. I think we’ll see people chasing the top models for quite a while, or indefinitely - depending on the cost curve.
It may not matter. Think about why SOTA model companies are exploring chips. What do chips offer?
If SOTA models haven’t peaked, then the SOTA model companies would still be churning out better and better intelligence.
Google rolled out TPUs in 2015. AWS released Inferentia and Trainium chips in 2020.
If companies working on ML-specific chips was evidence that large transformer models have fully saturated their potential, the field would have been done circa GPT-2.
Yeah this feels like Altman not knowing engineering well enough to realize where the focus should be.
Like he is optimizing to keep providing a vanilla token factory when weighted chips are coming and local models will supplement.
My head canon is savvy chip execs will be etching architecture his OpenAI pioneered into their flagship products while trying to minimize how much foothold he can get in hardware. Murica done offshored it. Not ours to control.
> I know this is what Taalas was doing (acquired by AMD), here was their demo, https://chatjimmy.ai/ which is based on Llama 3.1 8B. It feels like this should start to happen soon.
Taalas needed a giant chip (6nm) for an 8B model.
At best you could use a more advanced node to try to put a MoE model across several chips working together, but you can’t have GPT Sol size models on a single chip like that.
> Taalas needed a giant chip (6nm) for an 8B model.
You're phrasing it like it was kind of an inherent technical limitation with this kind of burning weights into silicon. Which is also not new, it goes back to the 1980s with fixed function digital signal processors and little linear regressions or hardware classifiers for industrial control systems, all are the same basic principle.
It's just usually not worth it to go super small process node, because most models people thought to turn into silicon were pretty small parameter sizes. We're talking 10-100 weight regression or at most 2-4k weight neural net, used in some instrument or factory equipment. You can do a decent MNIST OCR with a 4k weight neural net. For this, 180/130nm is fine.
Or you might think it's required with their special 4-bit as transistor thing (plausible). It's more that when you're experimenting and iterating, TSMC 6nm is their advertised path for rapid prototyping at cost for proof of concepts. And that's already in hot demand, while good luck if you're a startup trying to break in with 3/4nm as your first run.
Nope. But we are hitting some pretty impressive levels with 128B models.
The other thing is, a lot of the time, model performance is improved with more 'thinking' time.
The thinking time is just more tokens... but instead of say 1000 tokens or 10,000 tokens worth of thinking its 1,000,000... how does that improve model performance? Could a 128B model hit levels of GPT Sol?
Thinking generates a ton of tokens. These baked in chips tend to not have a lot of memory for context. I am not sure taalas supports Thinking at all.
The more problem like these they solve the more they will look like GPU.
couldn't one just add some hundreds of GB of HBM?
Yeah, but then there's the size of KV cache needing to be read through that HBM interface for each token, putting a hard limit on the tok/s based on the memory bandwidth.
On some models a large context can be a notable proportion of the size of the weights themselves.
For example, qwen 3.8 27b uses ~64kb/token for the kv cache - so for a 256k token context that's ~16gb of the kv cache for a ~54gb model (assuming 2 bytes-per-param/f16 for both).
So if the current non-baked-in chip is already memory bandwidth bound, as is often the case for current hardware and models, and the "only KV cache in HBM" chip has the same total memory bandwidth, it can only ever be (54/16)=~3.4x faster for the baked in-silicon model.
EDIT: I guess actually (54+16)/16=~4.3x faster, as the current implementation would need to read that KV cache too :)
Yes, they could also sell me GPT Sol 5.6 or 5.7 on a chip and I’d probably buy it. It’s a really really useful model for me, I’m not sure how much better for coding I need it to be. For most things I find Sol good enough with a small amount of coaxing around my tastes.
Keep in mind that what previous work has done on a single chip with weights baked in was on a 8b parameter model. Sol is likely something in the 5T parameter range, perhaps higher. Serving the whole thing at BF16 is on the order of $3m in hardware just to serve it at all, and closer to $1-1.5m of hardware if it was being served as NVFP4. And power draw starting at high tens to low hundreds of kilowatts.
Let's say a magic set of chips comes along to host this. Maybe it's 2-3x more efficient in size and power. You're still talking a form factor that's a good chunk of a rack, draws tens of kilowatts, and could actually be sold at a similar if not higher price point because the OPEX is so much lower.
It may be useful but it's certainly uneconomic to spend >$1m to self host the model, plus ongoing power and maintenance costs, plus the cost to adapt whatever building you're in to be able to power it.
ill give you that the way we talk about this ppl seem to think wed do this tomorrow, but in the 70s a kb of ram took an entire rack and tons of power also. Its seems equally plausible that we could go into a cycle of iterative refinement of baked model hardware that would end up in "personal ai" just like we got to personal computing.
Man wouldn’t it be cool to be able to slot a massive ROM AI chip into the external AI drive of the pc…
It should look like a NES cartridge! That you have to blow on its end to clear any dust and it should do a satisfying click when it slots in.
Cooling might be an issue though...
It'd just be pcie probably
Also right now Sol 5.6 Max is super slow but if it were way faster on a chip (like Taalas' Llama 8b demo) then it would be an extreme value multiplier. But the model is so large that "baking it onto a chip" doesn't seem straightforward.
That'd be ungodly expensive.
> For example, GPT Sol baked into a custom chip run for $100M that runs 10x as fast and 10x as cheap should pay for itself as long as the chip is useful for long enough.
but you trade updatability, which I don't think is worth it yet.
Maybe! (1) Would SOL level intelligence be useful 3 years from now? 5 years? (2) would dedicated chips be the most affordable way to run this model in 3-5 years?
I suspect the answer to both of these questions is yes right now, but I agree it’s borderline.
3 years is an eternity.
Eventually, someone is going to do this in Minecraft
My guess is we only see this once they start saturating computer use benchmarks. That's a use case which would be extremely valuable at the right costs/speed, but the current models just aren't there yet.
"Baking in" a model into a chip is a bad idea because chips take 2 years to tape out and then you're stuck doing inference on llama 3 in 2026 when fable/sol are available. Every accelerator is a tradeoff between flexibility and performance and GPUs are already pareto-optimal
It depends when the good enough level hits. Pretty sure we are almost there for most common applications of AI.
That's only half the problem. OpenAI is contractually obligated, if you will, to believe that models will continue improving at an impressive rate for the foreseeable future (otherwise their valuation makes no sense).
If you believe that, then you should expect to get Sol-level performance out of a Luna-cost model within six months or a year. If you have a system with the weights baked in, that means you're going to end up serving that Sol-class model several times more expensively than it will take someone who comes along a few months later. (such as what recently happened with DeepSeek's update.)
And under that assumption of continuing advancement, baking things in doesn't make sense in general - it's a play you'd make if you think things are slowing down a lot. Which may be right but it's not OpenAI or anthropic's play.
Assume a $800 billion valuation. $100 billion ad network. $30 billion op income. 26x price to op income ratio. It's right there for them to grab, or someone else to grab.
Their valuation does make sense if you believe: 1) they can retain a massive user base and 2) a massive user base can be monetized. Future value is almost always pulled forward these days for high growth tech companies.
An LLM the size of Google search in users is even more valuable than Google search. The ad market for LLMs will be even larger than search was (no matter what HN prefers).
The monetization part is the easier part. Silicon Valley understands extraordinarily well how to build ad networks. If OpenAI maintain their gigantic user base, a $100 billion ad network is a given bolt-on. They'd have to screw that up in an epic way to not get there.
Facebook - Insta - WhatsApp is an absolute dogshit tandem with a gigantic user base. $228 billion in ad sales and still expanding 10% per year.
Google knows this is what's happening, that's why they don't care about chasing Anthropic very much. They're busy completely remaking how their core search business works.
Good enough will hit when the tech stops advancing quickly. You could have a "good enough" model but in 2 years if the general purpose chip can run it just as fast, there is no point having the single purpose one.
The whole point is that it's supposed to be more efficient. But models are also still getting absurdly more efficient every year, so you're likely nullifying much/most of the advantage. 18 months is a long time right now (and 18 is only time to tape out, not operational in data centers).
Even if the balance was net positive, you would also not be able to train them against new tools/harnesses or knowledge. How many years do you expect to keep using them?
The good enough level isn't ever arriving. We're in the first or second inning for LLMs. They will rapidly subdivide in complexity, they will not stagnate in the next decade.
Beyond the model, when would you freeze processor performance, such that it was good enough? Because that's exactly what freezing on Talaas is premised around.
The semiconductor technology will also continue to improve. You lose twice. Talaas is one of the dumbest ideas I've seen in semiconductors in decades.
There are other options. I worked for a startup called NVXL and we were programming DNNs into FPGAs using OpenCL, on custom boards we built to plug into NVME. It worked great, but it wasn't fast enough at the time to compete with Nvidia, or even Intel AVX512. Ultimately the company failed, but maybe some hybrid like that could work for LLMs? I haven't been up to date on how DNNs and LLMs look like under the hood these days, but there's got to be someone doing something similar.
Well we'll see those surplus chips being repurposed for toys then. Who wouldn't want a new Furby that can actually hold a conversation.
I've seen several attempts even on HN of the LLM meets Teddy Ruxpin (or more accurately AG Talking Bear) but most of them offloaded the AI to some off-site servers.
I’d like to think that most parents would be weary of handing their children what basically amounts to a tape recorder that siphons all the data off to a large corporation.
OTOH, a completely local one (LLM + VAD + Speech Rec) would be a fun little thing to build.
https://en.wikipedia.org/wiki/AG_Bear
It won't happen until IPO. If they do it now it'd be signaling that AI isn't improving fast.
isn't that what they are doing with cerebras?
No, Cerebras holds the weights in SRAM - they are changeable, not baked in.
etched tried this.... it didn't go very well
I would assume asic based llm would work really well. Why did it not go well?
https://chatjimmy.ai/ runs Llama 3.1-8B on an ASIC as a demo by https://taalas.com/ I believe.
That's quite a few parameters shy of today's trillion-weight behemoths, but it is fast.
You are correct. I think this is the bull case. It seems like this would be useful right now for some things (eg moderation).
These nascent inference chip efforts are reminding me of the early 3dfx / riva / mach / powervr days. Will be interesting to see if inference chips are here to stay and, if so, who the eventual dominant player(s) will be
Which in turn reminds me of Soundblaster audio cards! I suspect inference chips are closer to the GPU story than the Soundblaster story though.
I remember one soundblaster card I bought came with a Lara Croft demo, that exploited the incredible immersion of real time dynamic reverb.
Genuinely I think game audio took a few steps back from that heady era, the innovation in audio likely didn't sell as many cards as graphics innovations did.
EAX was very powerful in its heyday, but it has died because of a thousand cuts.
First we had to have the audio processor. Good EAX was available on top of the line cards, and they were not always cheap. Lower end chips got less features.
Then we had to have the speaker setup to have the greatest sound, or needed to get a real 5.1 headphones, which were bulky and never provided the same fidelity.
Then Microsoft changed the Windows driver model, cutting the driver's direct access to the card. All of the timing sensitive effects were gone in an instant. I remember installing the new drivers and getting literally nothing. Sound Blaster was the only card with an hardware mixer, and Microsoft didn't feel like enabling them. Mixing at the DirectX layer killed the cards.
Soundblaster's very closed stance didn't help them either. None of the cards after Audigy2 worked with Linux when I had my desktop system.
After my Audigy2ZS, I moved to Asus Xonar D2X. Its positional audio capabilities were nice, but I mostly bought it for its Linux support and sound quality, and that was top notch in that regards.
Then sound cards became commodity. Everybody stopped making good cards. Musicians moved to audio interfaces, audiophiles moved to DACs.
Just looked to the SoundBlaster website. Internal cards are very limited. One DAC, one DTS enabled 7.1 sound card for PC cinema systems, three game oriented lower end cards, nothing else.
The DSPs that could double as a sound card and “soft” modem ruined their market in short order.
Eeeeeh idk about the Sound Blaster comparison. Creative earned their place in the early-mid 90s solely because they were the one company making a sound card with drivers that actually worked properly.
It wasn't really the cool reverb effects or wave tables, though those were a nice bonus. It was just "I can tell my computer to make sound and it actually makes sound without days of troubleshooting."
Granted, similar things could be said about 3dfx. It's was a 3D card with drivers that actually worked.
And then there's the obvious "sound blasters and voodoos go in my computer, jalapeno goes in someone else's computer" thing.
On board got "good enough" and the separate cards died away.
In fairness on board (depending on the board but on the whole) is pretty good.
Every company is designing their own chips so the dominant players will be one level down: Broadcom, TSMC, Hynix/Samsung/Micron, etc.
There is drastically more power and profit in the software ultimately.
Apple is in the software first, the hardware second. Everyone at Apple has been trained to understand this for decades, and Jobs pointed it out endlessly. Apple's real moat is software (services, iOS, experience, MacOS).
Windows, Office, Azure, et al. Microsoft accumulated approximately one zillion dollars in profit on the back of software. It's a vastly superior business to anything hardware has traditionally seen. Nvidia is the first true juggernaut hardware profit machine, and the AI boom in extended hardware (RAM, storage) will prove temporary (even if there is a feast during that time). Microsoft's advantage and moat was Windows-Office for decades. It was a far better business than Intel's chip biz.
Google is a software company first. Every aspect of what made them and maintains them is software first, hardware second. They're a $400 billion software company. Their ad machine is software. Search is software.
Facebook is software. Instagram is software. WhatsApp is software. A $200 billion software company. They're not selling hardware, they're selling ads via software, they're monetizing users that use their software.
AWS is at least half software as an entity in terms of complexity, competitive advantage, et al. That's a two trillion dollar business.
LLMs can run successfully with various hardware approaches. The software is the value at the end of this, regardless of the hardware under it. The sole exception so far that may be sustainable is Nvidia, and we'll see if the bottom falls out from under that margin monster (China, specialized AI chips, whatever it happens to be that cuts under them massively).
Hardware always gets its margin squeezed eventually because it's a manufactured good (with inventory, fabs, etc). Software is hyper margin by default, you have to layer a lot of garbage on top of it to kill the margin. Nvidia is 33 years old, they have had a rich business for three years, that's it.
The AI boom is the sole reason anything in hardware has looked great in the past 20 years. Check the margins & op income for the top 20 hardware companies, from TI to AMD to Intel to Nvidia to Micron to Sandisk to Samsung to TSMC to ASML, prior to the AI boom of the past couple years. It won't last indefinitely. And after the return to a more normal environment happens, the hyper margins in software will persist.
> Will be interesting to see if inference chips are here to stay
To me, the efficiency gains of inference chips are so significant that they are certainly here to stay — barring a revolution of sorts that leads to a world devoid of AI as we know it.
> who the eventual dominant player(s) will be
This couldn't have been easy. The team at OpenAI has worked a miracle.
I bet cost is of no issue with the capx where it is at. It is almost certainly organizational. Meta throws money at every problem and it never seems to workout for them.
It's so funny to see FP4.... I remember 20 years ago being asked what sort of HPC we needed in genomics, and the answer was basically, "lower precision, faster" for the stuff I was working on. But FP4 is, well, almost comical.
One thing not on that comparison table: die size. If I'm understanding that correctly, it's about the same as the Rubin, but at 1/3 the number of NVFP4 PFLOPs. (The text disagrees with the table, I'm taking the table as truth, perhaps that's wrong...)
Agree, I remember when even half precision made its way into C# sometime around 2020 (I didn’t know much about ML then) and I thought, well I guess that’s a worthwhile tradeoff but I can’t imagine going lower. Lo and behold (1-bit Bonsai) how much lower you could go.
Ternary?
Knuth's base-e proposal enters the chat.
They were right about everything 50+ years ago, but they didn't have the budget for the right hardware, had to write conference papers and books instead.
I can totally see how ternary would work from a physical implementation perspective but I have a really hard time visualizing anything using base-e, can you explain how such a thing would work in practice?
No it's impossible. But it would be optimal!
Ah, the spherical cow of number bases :) Thanks for the response, that saved me a sleepless night.
I hadn't seen the token/Joules comparison with human speech before. Humans are still 22x more efficient, which is not that far considering the rate of progress in this area.
The 20W number includes EVERYTHING else the brain does. The chips/models are literally only producing tokens. Let's see an LLM drive a robot harness and have the robot produce speech, as well as move through 3D space, keep track of metabolic needs, etc. etc. etc. before we compare efficiencies. That is even assuming the tokens are of equal quality. This comparison is currently Apples and Oranges.
kind of a moot point if you can't get your brain to not do everything else. I think it's a fun comparison, even if it's not a 100% equivalence.
Right, I can do the talked about ~3 tok/sec output and drive a car, hold my bladder, and eat chips at the same time.
Take that, Jalapeno!
For what it's worth, LLMs don't really suffer from incontinence, so at least that part is pretty much a solved problem.
They sometimes leak their system prompt
So why are their water cooling systems filled with leak detectors? /s
And the brain is literally only producing electrochemical signals.
I don’t see how tokens can’t produce speech or track metabolic needs. You can talk to chatgpt can’t you? Or do you mean literally talking? Because that’s not a brain function, that’s the mouth, vocal chords, and lungs.
> I don’t see how tokens can’t produce speech or track metabolic needs.
It probably could, but the point is this would require additional tokens, blowing up the comparison. The token output of LLMs and "token output" of speech are simply at different abstraction levels. Hence my comparison to the LLM brain driving the robot harness to produce speech etc. This would be more comparable, and also look significantly worse than "only" the 22x less efficient number.
22 times more efficient is not like 22 times more powerful. It’s extremely harder to close the gap in power efficiency than in raw power. Simply because the power efficiency we see is the result of billion years evolution optimization.
But the true number is IMO far bigger: orders of magnitude greater if we think in terms of equivalent performance.
I couldn't source the parameters from the screenshot or the nearby graphs, but from the nearby graphs you can see that at concurrency C=1, tokens/Joule (vertical axis) has totally plummeted, and obviously concurrent inference is much more efficient by batching. Divide the memory by the bandwidth and thats how long it takes to dump the full RAM contents through the chip. Do you want to do this once per token for a single conversation, or do you want to progress multiple conversations if you're going through all the weights anyway? The peak in the graphs is easily 22x more efficient than the low bottom right part on the graphs. So in batched mode its already more efficient than human speech.
> Humans are still 22x more efficient, which is not that far considering the rate of progress in this area.
Based on a human output rate of 3.3 tok/s, which seems questionable as a means of comparison
I do believe that this is the trade off. We are more efficient but slower in terms of thinking (at the same level of intelligence). Some animals go much further in terms of that trade off, see https://en.wikipedia.org/wiki/Portia_(spider) for example.
Just checked wikipedia page... they have like 100K neurons only.. wtf!!! How can nature cramp all senses, including spatial, motion, life maintenance and general thinking into just 100K neurons?
What exactly are you questioning?
The claim that tok/s independent of quality is a useful comparison (I can get thousands of tok/s on a suitable small model), and secondarily that humans can’t output “tokens” faster than than in some sense, which I am less confident about
At just inference! Which both a human and a model can not do without training, but while training rounds to zero for the model, for humans it scales linearly.
I am relatively certain we have already squarely been beaten in net efficiency at scale.
Probably not when you consider the training cost and upkeep expenses, not to mention the depreciation…
I wonder how that stacks up if you consider all the time you have to keep the body alive when it’s not actively producing “tokens”.
Careful, let's not put the whole matrix into stasis outside of business hours.
Productivity is not the only reason to let these meatbags burn oxygen.
Fairly amazing when you think about it, like human intellect can run on a bowl of rice and a chicken yakitori skewer.
You’re missing the factor for intelligence/token.
I mean, surely when quality is accounted for the difference is significantly higher
Or maybe significantly lower.
I love how now you have to consider the possible s** posting motivation behind analysis of a trillion dollar industry being conducted at a world-class level by a bunch of ex Reddit and 4Chan adjacent mods -- it's one of the best stories in AI that SemiAnalysis is not cut from the same cloth as Gartner McKinsey et al
SemiAnalysis’ founder was roommates with Anthropic people, not OpenAI, so he may be slightly (very slightly) more objective here.
Along with Leopold Aschenbrenner so maybe not so much.
The guy that was part of FTX, fired from openai for alleged theft, got billions in a hedge fund somehow then lost billions. Why are all these people so scummy? It is like voting Trump three times in a row.
They're very intelligent people who do very clever things at a young age, which draws the attention of very rich people who can exploit them to get richer, and no one tells the young person they're being exploited. They're heaped with praise and 'wealth' (millions, but crumbs compared to what they're making for other people), and told they're geniuses who can do no wrong, mostly by the media that happens to be owned by the rich.
Then the rich people pull the rug leaving them holding the bag, and they move on to the next young clever group.
And the cycle continues.
s** posting? sex posting?
shit posting
Thats what I thought too but then it would be s**?
i see 'hunter2'
s**?
Edit: OK, hn is removing one *
If it's trying to convert it to italics, you may have to use a backslash to escape them
Or double them up: s****** gives s***.
I like that to type s****** you had to type s************.
Or escape them;)
Why censor yourself?
Bots do that because other platforms remove or hide posts with bad words
Humans do it because they've been raised not to swear.
The "industry news and research" part of the AI industry feels very... suspect to me. My intuition is telling me that it's a bunch of people with influencer-y type social media skills and no actual credentials just grifting because there's so much money floating around.
What credentials do you need to write a substack about an industry so it’s not grifting?
"not cut from the same cloth as Gartner McKinsey et al"
Yeah, those guys aren't biased at all.
The semianalysis people have scripts which incorrectly count their numerators and denominators all the time. All their benchmarks are flawed. It is such a slipshod operation and they charge exorbitant amounts of money for it.
Say more about this please
semianalysis is pretty good
Are they? https://jon4hotaisle.substack.com/p/influence-as-a-service-s...
> I love how now you have to consider the possible s*** posting motivation behind analysis of a trillion dollar industry being conducted at a world-class level by a bunch of ex Reddit and 4Chan adjacent mods
I mean, previously you could have said something much the same except substitute "frat boys".
> McKinsey
lol. lmao even.
Have you seen the quality of their output? I'd take Claude or ChatGPT Free Tier over advice from McKinsey these days.
Continued hardware improvements really make it hard for me to believe token prices will not continue to plummet.
This may just be a classic case of Jevons paradox: https://en.wikipedia.org/wiki/Jevons_paradox
In short, better hardware will drive down token cost in the near-term, but will drive up the demand for tokens as it gets cheap enough for other sectors to start to use it heavily.
It comes from steam engines where economists originally thought that coal demand would plummet with more efficient engines, but it actually just meant that we found more uses for steam engines.
This is exactly what I see happening now.
Codex keeps doing these usage resets. What do I do? Burn even more tokens than ever before. I know I'm not the only one.
Is this a normal thing now? I remember seeing this talked about as a surprising thing, but now I'm seeing posts about this as if it's normal.
(I switched to using local models as usage limits, api instability and the concept of paying per token stresses me out)
If we are applying Jevons paradox to this then the unit being consumed is not tokens but the inputs for token production - power, capex, something else. To draw an analogy to the steam engine, coal:electricity::mechanical-work:tokens. Jevons paradox does not talk about mechanical work becoming cheaper in the short term setting up a sort of rubber band of demand creating spiking prices for mechanical work. Compared to the renaissance, mechanical work was much cheaper throughout the industrial revolution and remains cheaper to this day. We can still definitely say that the easier it is to produce tokens, the cheaper they will be.
All Jevon’s paradox says is that as a resource becomes cheaper total consumption of that resource increases. It applies equally well to the inputs of token production as it does to the tokens themselves. The former would describe the effect the sellers into AI companies see (energy, GPU chips, RAM etc - if they lower their prices they’ll have more overall consumption) while the latter describes what the AI companies see with their customers (if they lower token prices consumers will use more tokens overall).
the total cost spent on tokens may go up, but i just cant imagine per token costs going up
Depends on compute capacity. If we become supply constrained on tokens, then prices will necessarily go up.
no they dont because inference stacks are getting more efficient and models are getting more intelligent per parameter.
I would posit there’s no way in hell they’re getting sufficiently cheaper on a short enough time frame vs how much demand is sky rocketing. AI companies are seeing quarterly doubling of revenue if not more.
It is incredibly cheap now. What sectors are you thinking of?
That's when demand is higher than capacity. Now imagine places like Gigalab and Chinese labs are online and able to produce significant percentage of chips. That could cause real surge in prices.
I think you're reducing a very complex thing (the global economy) into a very simplistic model (Jevons' paradox) and thinking both are the same thing. This has no predictive power or rigor. You're just wishing things would happen as they did before, without considering that conditions and situations change significantly, and instead of Jevon's paradox, we look back at today 50 years from now and talk about Jensen's paradox.
This doesn't mean the concept is BS, but one single concept cannot explain away everything in such a system.
There is just so much downward pressure on token price, from every direction. We would need a completely new understanding of economics to explain why the price shouldn’t go down. Or market collusion/regulatory manipulation.
The demand for them is growing _per person_, not just across the wider economy, if tokens cost half as much but you want to use 3 times as much you're going to have to pay more.
Maybe 1000s of tokens per second unlocks realtime robotic decision making, and now every robot needs to continuously stream tokens to and from the cloud to operate. That could 1000x demand overnight, just to speculate :)
I would very much like it if anything that moves with appreciable mass is governed locally just in case the link drops and/or latency suddenly goes up. Motion is very unforgiving and accidents will happen if that's not taken into account.
I think you just found what we will see in the S-1 prospectus of OpenAI
Seems unsafe to make locomotive decisions remotely
Think about the agents buying computers for their agents. /s
The price has been going down for ages, its not clear what you are pointing at
Pointing at the nay sayers who say tokens are heavily subsidized and it’s all going to come crashing down soon, surely any moment now
I mean, it will obviously crash at some point. With so much pressure on token price to go down that means way less opportunity for margin for AI providers. OpenAI is in a pretty bad situation
What does this have to do with margins? It can remain the same once prices go down
At the price going down? And that it will continue to go down, even if the hardware improvements stop. Not sure what isn’t clear
Token prices coming down means nothing if the models keep wasting them
With the corollary that old hardware valuations will plummet with them.
Although given we have marginal pricing we need to push through to those lower prices in the face of increasing demand, so timing of this is uncertain and the key to the AI financial markets
Hopefully this also means billionaires can stop trying to drop data centers into residential neighborhoods with zero noise control and polluting on-site generators, signing local politicians on with NDAs, calling for eminent domain to seize homes to build power lines to data centers, etc. etc. etc. Not to mention the water use controversy.
Token prices plummeting is probably a good thing, but not without the regulatory backstops that prevent these effectively industrial facilities from being operated with no regard for the externalities they impose on people who live near them.
Nah, Jevon’s Paradox says that cheaper tokens will mean increased overall energy consumption.
If we can’t even build data centers, the least disruptive industrial use possible, there’s no hope to reindustrialize the US or anywhere outside of China.
>polluting on-site generators
how much pollution do you believe modern gas-turbine engines to produce?
>Not to mention the water use controversy.
what percentage of US water usage do you believe is by AI data centers?
Yeah but is it really even as good as Rubin? Seems just competitive.
this is a story about a proprietary accelerator being built/designed by a token provider. and you think they're going to return the efficiency gains to the customer instead of capture the value for themselves? interesting take.
OpenAI just dropped the price of Luna by 80% and Sol by 20-30%
and amazon shipping used to be free without prime, and uber used to be cheaper than taxis, and airbnb used to be cheaper than hotels.
you really don't get it?
almost every pure tech commodity has gone down in price
- gpus
- retail computers
- laptops
- ~gpu~ appliances like washing machines
- cloud computing
i think you don't get how economy usually works in tech
I'm especially enjoying how RAM and SSDs are going down in price.
GPUs and memory have gone up in price. It's more expensive to buy a 1-2 year old video card than it was at launch, sometimes by a shockingly large factor. Laptop vendors have recently shipped flagship models with less memory than the previous model, because they can't match price expectations for a laptop.
I was checking laptops today for an upgrade from the model I bought back in 2019 and it's not gonna happen from how cheap they are.
listing gpu's here is crazy considering the current prices
GPUs and laptops and memory and storage are all crazy expensive
i figured out why this comment is so confusing: this is actually a message from the past, around 2020. either that or simianwords is a time traveler that arrived today and hasn't read the news yet.
We should be mindful of the context that many of these providers VERY likely have been selling their subscriptions at a substantial loss
So as much as i agree “more profits to stakeholders screw the customer”, i think its more of an emergency to get to profitability before the music stops.
> We should be mindful of the context that many of these providers VERY likely have been selling their subscriptions at a substantial loss.
what makes you think this?
Because everyone keeps saying this so it must be true. Real "it is known" kind of vibe with these statements.
He's a subscription truther. There's loads of them. OpenAI's profit increases with each subscription that is cancelled. Pretty soon they'll have more profit than God.
Yes, I can bet on this happening. If anything, this is a net gain for consumers as it is a competitive market.
go ahead and bet: alibaba is a publicly traded company
This is most impressive. The interesting question to me, is outside of the LLM accelerator space: will generalized chips have massive leaps in performance once LLM technology is used to create the next generation? In general, will we see rapid advances while we extract the value of these models in creating architectures? I'm so far removed from the space that this is a very naive interpretation of all this, but I'm curious.
ASICs always do better than general purpose chips. General purpose chips is turtles and turtles of virtualization and have to consider 4+ decades of backward compatible instructions set support.
ASICs are deployed when the application area is economically large enough to so there is return on the investment in developing one. Bitcoin mining few years ago or today inference or more mundane things like video decoding/encoding.
General purpose chips on the other hand have to be general purpose first to be useful, i.e. support as many application domains and instruction sets as possible . It can be long tail of support which both slow your chip down and also slow development down. Apple's took a long time to develop M series to be general purpose enough and still need even now software tooling like Rosetta to make say virtualization work for a good reason.
New tooling would always help and there is already lot of software emulation for developing chips today but you still need physical iterations to tap-out and have high enough yield, no LLM can help with that.
Existing CPUs have been extremely optimized by ~6 competing, well-funded teams. I expect AI to accelerate things somewhat but it's not clear that there is any low-hanging fruit available for AI to find.
All of these words spilled and no mention of the ISA.
That's because it's AI-slopped.
I don’t think this is public?
The reliance on Deepseek and Kimi as the benchmarks from every chip maker from NVIDIA to OpenAI is a good tell of where things are heading. In the next couple of years, hopefully we will have systems at home for everyday use and corporations can buy bulk from providers.
As opposed to closed-source models? Benchmarks for GPT Sol wouldn’t be particularly meaningful, as no one else can run the benchmark, and we don’t know what the exact model specs are.
Picking the best open source models is really the best they can do.
This means that they're going to want to IPO soon - this is good news for investors + they need the capital.
No, this is because they want to IPO soon.
If the chips weren't this compelling they would have something different to announce.
These are paperclip maximizers who just happen to wear human skin - there is no underlying premise nor ideological goal.
So this will make GPUs and associated affordable for people, rigjt ?
No, it's the wafer starts that are driving prices. Switching from Nvidia to custom doesn't change the constraints.
Competition is good for all of us, we will get better and faster chips.
Or at least Nvidia GPUs will become slightly cheaper for regular consumers again
That's if any datacenters are allowed to be built with them.
There is probably a ~50% chance that the next Dem candidate for presidency runs on a national datacenter moratorium or something equally as crippling.
If the populist campaign is to Make Affordable DRAM Again, then it's not a terrible solution.
The current datacenter owners love a compute-bound world anyhow. A moratorium on new datacenters would increase their valuation, encourage efficiency and make computers cheap again. If Chinese labs can ship frontier models under 1T parameters, why not American labs too?
No way to avoid the memory cartel, even if CXMT catches up.
These are not replacing GPUs, they are entirely complementary. It's the same with cerebras, groq etc, they are all complementary to the GPU.
If you think tanking Trillions in investments, warming the earth and increasion ocean water levels, creating water shortages and brown-outs is "good for all of us" - well, the rest of us beg to differ.
these GPUs make computation faster, I understand as of now maybe all the computation is used to generate yet another junk LinkedIn post or unnecessary RFC, but at some point this craze should settle and we will be left with powerful computation machines, which can be used for computing more useful things
The GPUs being paid for w/ billions in investment will be obsolete and e-waste in a few short years same as a Cray-2 was just a decade after its release.
It's fine if you're one of the people selling shovels to gold miners for a while, but sucks to be building houses in the boom town?
The pricing of GPUs themselves aren't really the problem: it's the VRAM that comes with them.
Sounds like a great way to get deals out of Nvidia.
Sure, but even heavily discounted Nvidia chips won’t be competitive for inference if they’re worse on perf/W.
Any article that features Sam's fucking creepy face should be marked with a jumpscare warning
I guess special hardware is the new moat in AI.
Maybe the money will still flow into this industry after all
I hope the LLM wave will leave GPUs behind to go back to pursue more general-purpose computation rather than spending their die area on multiplying 4-bit-number matrices and such things.
Is that not literally the exect opposite of the direction asics for LLM inference is going?
Is this bad news for Cerebras?
When people talk about the commodification of inferencing, they imagine a future where everyone has access to frontier models and can run them at the same cost, and what will actually happen is closer to the commodification of _oil_, where only a few companies have the scale to produce it at a competitive price, and advances like this are _why_.
Once models are more or less interchangeable, the price of LLMs will drop to essentially the price of energy required to run them, and the big labs will be able to run them cheaper than anyone else.
I don't agree. At the moment companies like NVIDIA take several times what it costs to make a chip. I think the fair split for the technology contribution is more like 50-50, maybe even 30-70 in favour of the manufacturer.
With competition we will actually have the fair split, whatever that is, and thus much lower prices.
At the moment, to have a big AI firm, or really AI firm at all, you need to be blessed by NVIDIA, in the form of receiving circular financing for your compute. They know that their prices aren't fair, or competitive.
Commoditization of inference is the end of that. The end of the mega-premium on inference hardware, and it's good not only for people who like running their LLMs, but it's the first step towards commoditization of training.
10-90 is the fair split.
I don't believe models will be commodified because each model is unique with strengths and weaknesses. Its not like Steel which is more or less the same no matter where you purchase it from.
If what you said were true, you would hardly see people complaining about the quality of Opus 5 or good writing from Sol. But people do.
The same level of intelligence gets roughly 10x cheaper per year. So you might both be correct where a large part are commodity tasks but frontier is hard and valuable and not commodities.
> I don't believe models will be commodified because each model is unique with strengths and weaknesses.
They are all converging.
This depends heavily on what the use-case is. Yes, if it's a coder making software and having to read LLM output then writing style matters. If the LLM is used in an automated data processing pipeline with a capped level of complexity, entirely different aspects matter and LLMs become more interchangeable.
> Its not like Steel which is more or less the same no matter where you purchase it from.
I’m not an expert in metallurgy by any means, but this seems really off. There are many recipes for steel and varied processes that also impact the final product.
The article is a bit naive:
> However, as previously mentioned, Jalapeño’s results are obtained without speculative decoding and Vera Rubin’s results use speculative decoding. Speculative decoding leads to a ~3-5x reduction in cost per token. When speculative decoding is implemented on Jalapeño, this will enable Jalapeño to serve tokens even more cost effectively.
How much speculative decoding improves throughput is workload-dependent. Yes, it can improve performance by 5x, but it can also slow down performance by 2x. It also requires a vastly more complex decode loop than the standard token-by-token decode. The tech journos didn't ask themselves if speculative decoding improves perf so much why wasn't it on by default? Perhaps, because speculative decoding is not optimal for OpenAI's ASIC? Perhaps that is also why they were only able to benchmark the not-very-representative single-user-decode case?
R@6.....111
hacker
Warning, this is a long comment! (I’m trying to stick to sourced facts here and not overstate what they mean)
I went down a rabbit hole after watching Dylan Patel on Dwarkesh today: https://www.youtube.com/watch?v=aV26V1UvkJw
I was initially just surprised by how bullish Dylan is on OpenAI/Anthropic and how bearish he is on China, despite Chinese labs getting closer to US SOTA while offering inference at dramatically lower prices.
So, I started digging while waiting for various day-job inference calls to return, ha.
Dylan says he spent years obsessively posting on hardware forums, moderating hardware subreddits, and running anonymous hardware blogs/videos before SemiAnalysis. But he also says most of that history is now gone, including from the Internet Archive, because he asked for it to be removed.[1]
In a 2024 interview he described his post-college job as “data science” around hurricane/earthquake/wildfire simulations for a financial company.[1] In a 2026 Sequoia interview he described himself as having been a “quant at a small quant risk firm” who generated $10M+ of “risk-free revenue.”[2] The Information reports that he declined to identify the employer and doesn’t list it on LinkedIn.[3]
Even harmless/silly stuff seems to drift. In February he said he kept bees for ~1.5 years. Today it was “few months, few months.”[4][5] I know, sort of silly and doesn't matter.
The Information reports that Patel owns stakes in ~20 startups in the same ecosystem SemiAnalysis covers, organized a $50M Fluidstack SPV, and is now targeting a $400M venture fund.[3][6]
And, in a 2022 HN discussion about SemiAnalysis disclosures, after saying his reports had moved smaller stocks by 20% in a day, Patel wrote: “If I thought I could move the stock, I'd make the position in the morning alongside my clients, and publish shortly after.”[7]
I don’t know that any of this is false or that anything improper happened (I’m definitely not claiming that). More that 1-2 of these things would just be odd. Taken together, though, they made me question how much trust I was putting in the broader story.
The dynamic of reminds me of crypto, WeWork, Theranos, Citron, etc. Once enough important people validate someone, things that would normally invite basic diligence somehow stop getting questioned.
[1] https://www.dwarkesh.com/p/dylan-jon
[2] https://sequoiacap.com/podcast/dylan-patel-of-semianalysis-w...
[3] https://www.theinformation.com/articles/dylan-patel-semianal...
[4] https://www.latent.space/p/dylanpatel-cooking
[5] https://www.dwarkesh.com/p/dylan-patel-3
[6] https://www.theinformation.com/briefings/exclusive-semianaly...
[7] https://news.ycombinator.com/item?id=31065646
Not publicly acknowledging how misallocation of capital may be happening today shows he is corrupt. He’s not that dumb to not know it’s a major risk to the whole story, and is certainly financially incentivised to write as he does.
If you research the origins of Dwarkesh, even more conspiracy level points emerge. I do not think even in the handful videos post Leopold fund collapse he has addressed it in any ways. That is the point of so called observers, they pretend to be impartial but everyone is a hustler in some way.
I genuinely curious who’s downvoting me and why. I do not understand this forum sometimes.
Once again we see the classic PR hype machine tactic of comparing a newer chip which is only available as an engineering sample to other chip designs which are widely available and much older.
They also fawn over the chip’s TDP when all other chips have to support 16 bit floating point and thus must run much hotter.
They make the classic mistake of equating max TDP with in-use-watts, and praise this magnificent (fictitious) performance per watt at FP8 with other chips’ max-TDP at FP16, which draw twice the power.
Evidence that the IPO can’t be far away.
Story says they're power limited. That's half-true. Actually they're water-limited. To generate power, you need water. To cool chips, you need water. If you try to use less water on one side, you need more water on the other side (it's physics ya'll, making and using energy generates heat which requires dissipation). The world's freshwater is diminishing while also being consumed at an alarming rate. The future AI oligarchs are whoever controls the most water.
The other side of the conversation is the idea that large models in DCs on custom silicon is the future. Maybe for enterprise? But consumers will eventually (10 yrs) have affordable hardware designed to run crazy-good local models (more RAM + higher bandwidth). That will take pressure off of datacenters, but also reduce AI profits, and move that money to consumer chip/device makers. Apple is once again the biggest winner. Nvidia consumer chips might get cheaper, but nerfed, to encourage datacenter use where they make more money. I'm hoping AMD can stop being terrible at software so that when we finally have their better hardware we can actually use it.
For datacenters specifically I've never understood what specifically consumes the water. Arent the water-cooling loops closed, so the water just cycles around and around and around?
They evaporate the water which is what makes it cool so effeciently.
Evaporative cooling does not necessitate an open loop system
The system which runs coolant over the chips can be closed but the part which uses an evaporative system to cool that is still open loop and vents water into the air, no?
Nope, it doesn't have to be open loop!
Example of such a system being used specifically for datacenters: https://blog.vantage-dc.com/2026/04/22/cooling-without-the-d...
Evaporated water is condensed, and in the process transfers its heat into another place that removes it. Another simple example is a pot of boiling water with a lid on it.
The link you cited is not evaporative cooling and a pot of boiling water with a sealed lid on it is a pressure vessel which eventually explodes.
If you were to remove the heat at a sufficient rate by, say, turning the lid into a heat exchanger, you would have a stable system.
How is that different from not using evaporative cooling and just putting the heat exchanger on the burner?
Yes they are for water cooling.
At the datacenter side, it depends on the method of cooling. You can chill the air or the chips directly (or both), doesn't matter, you still need to cool, and that still needs water. The question is, where is the water being used?
- If they use either evaporative cooling or a liquid-cooled heat exchanger, that uses tons of water consistently. This requires less energy (it's mostly passive) so you use more water.
- If they use closed-loop water cooling and/or heat pumps/electric chillers, that uses much less water - at the DC. But it does require more energy to circulate the water, run fans, etc. If you are using more energy, where is the energy coming from? It's coming from power plants, which require... you guessed it... more water (e.g. thermoelectric, hydroelectric, geothermal, concentrated solar). They need water in order to generate the power, and lots of it. Coal, natural gas, nuclear, and concentrated solar, all use steam to generate energy. Nuclear also uses water to cool the reactor. And water is used extensively to extract coal, oil, and natural gas. Geothermal uses water in the ground.
You can't not use a ton of water in one fashion or another. It just depends what method, and on what end the water is used. And the crazy thing is, most new datacenters are being built in places with extremely little water. Guess how that's gonna work out as the planet gets hotter?
I don't know why I got downvoted to hell for stating facts every datacenter architect knows. HN be HN'in.
This only makes sense if you never looked at comparative water usage rates and available water.
How can OpenAI mass produce this chip at scale more economically than Nvidia which has experience in the supply chain and scale efficiencies to do it efficiently?
NVidia has enormous operating margins, so a competitive solution doesn't have to match or beat NVidia's scale efficiencies; it just has to beat delivered cost.
One objective of the project might be simply to provide credible negotiating leverage when dealing with existing suppliers like NVidia. You don't have to deploy at scale for that to work, but you do have to look like you could if pushed hard enough.
> NVidia has enormous operating margins, so a competitive solution doesn't have to match or beat NVidia's scale efficiencies; it just has to beat delivered cost.
But then that means you have no actual moat against the behemot, right? Your competitor can move into the market as soon as they want to, at much better cost (so at slightly better price)... and Nvidia certainly can adapt much faster around hard hardware specs innovation than a new entrant ever could.
Those are not OpenAI's concerns - they just need to scare NVidia enough to lower their prices more than they'd otherwise want.
But they are also PURCHASING from nvidia so any time nvidia lowers their prices they save money.
In the short and medium term, it probably won't be more economical to produce for OpenAI. Where OpenAI is benefitting from their own chip is being able to tailor it to their models and workloads. When you buy off the shelf Nvidia, its not perfectly tailored and OpenAI has to spend marginally more to run off that chip. At the scale OpenAI is operating at and plans to operate at, that margin becomes pretty big $$
Because it's actually Broadcom that is doing most of the work.
By leveraging the experience Broadcom has in this area. Still remains to be seen how that goes when they want to scale production.
Replace OpenAI with Apple and Nvidia with Intel.
"How can OpenAI produce a LLM at scale more economically than Google, Amazon, and Microsoft which have experience in planet scale software and scale efficiencies unlike them".
One answer is they're quite good at poaching talent.
WARNING AI HYPE
Why they don't research how to make their own RAM and they have to buy it from the common market?
They should GTFO with this crap.
Create barriers to computing for ordinary people while milking businesses for tokens.
Building a custom-designed ASIC is much easier than producing state of the art memory chips.
There's a reason why Micron and Nvidia are the crown jewels of American technology right now and for the foreseeable future.
Nvidia buys the memory it uses on its GPUs, same as all other ASICs.
To give some context, Intel started making DRAM, I think they were actually the company that came up with modern memory techniques. They exited the market and pursued a more lucrative moat with CPUs.
Nvidia does not make RAM
That doesn't excuse them from wrecking the market for ordinary person.
NVIDIA produces memory?
Fabless AFAIK. And that's the actual problem - drawing up CAD diagrams doesn't help if the factories are fully booked out.
A state of the art GPU is much harder to design & produce at scale and than an internal ASIC.
People keep saying stuff like this without understanding what it takes to make RAM. It's one of, if not the most, heavily patented things in the world. The second you dip your toes into those waters the lawsuits begin.
If somehow you get around the patent issues, you're now faced with huge research and development costs, fabs to build, processes to sort out and all of that has very high failure rates.
Last time I checked Micron was the largest patent holder in the world and even for them this is a hard area where they are number 3 in the market.
RAM chips are not hard to produce compared to many other types of semiconductors; Intel started in the memory game and left because the margins weren't great and they were going to fold. The failure rates on these chips are actually very tolerable; you can have a very bad yield and still have a viable chip due to things like ECC.
Intel entered the memory space because they partnered with Micron. They left the memory space when Micron pulled out of the partnership.
Intel started making DRAM in 1970, Micron was founded in 1978.
Yes, it is difficult, but shafting working class is easy, therefor it is okay.
If the rich decided to buy all drinking water, you would probably be saying that's okay, making water is difficult, shortly before dying.
Well Sam Altman finally has built a moat against Chinese open weight AI. Well done. But what will this mean for Cerebras?
I remember when Tesla was building its own inference chips, and after about 2 years and billions spent, the whole effort was scuttled b/c they simply could not keep up with the iteration and R&D cycles of dedicated chip companies. I suspect the same will be the case with OpenAI vs Cerebras + Nvidia/Groq
Cerebras is targeting a distinctly different point on the cost/latency curve. They are betting that there will be some high value applications where latency and not just throughput is super important.
It is being used as part of a combined system. For example AWS is pushing for Trainium + WSE 3. The WSE 3 does the decode and the Trainium does the prefill.
Even in nvidia land rubin + LPU does a similar thing.
It has its downsides of course - if your traffic swings prefill heavy to decode heavy, you can't suddenly use your lpu for prefill. With GPUs they're totally interchangeable. Tradeoffs.
AFAIK You can use WSE/LPU for prefill, it's just less efficient to do so.
> Well Sam Altman finally has built a moat against Chinese open weight AI
Hes got a press release.
The issue is, baking something to silicon requires discipline and about 2 years.
This isn't something you can just change your mind on halfway through. Trust me, I know. You need a clear vision of what you want to support, why and what bits of a chip you need to achieve that.
And yet, the top comment is about “hardcoding” weights into the silicon.
Man, if only someone made like, chips that could lots of different calculations all at the same time!
I think the Chinese are going to be building their own chips aided with AI. DeepSeek, z.AI, MiniMax, Moonshot, etc, it's a race. The take off has really started.
> and after about 2 years and billions spent, the whole effort was scuttled b/c they simply could not keep up with the iteration and R&D cycles of dedicated chip companies
That sounds quite like...nonsense?
Chip companies work on years-long cycles. They know today what are they launching 4-5 years from now.